[vm/compiler] ARM64: Block R22 to hold NullObject().

Block R22 to hold a cached version of NullObject();
refresh it in the same way as for BARRIER_MASK register.
Alter assembler to avoid emitting load and to use NR
register directly instead.

This change improves Flutter Gallery code size by:

Instr: -1.81%.
Total: -1.16%.

Change-Id: Ifec654e799737527eec1d8a0e87b4a197ad0298a
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/125662
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>

runtime/vm/compiler/assembler/assembler_arm64.cc

@@ -475,7 +475,9 @@ void Assembler::LoadObjectHelper(Register dst,
                                  bool is_unique) {
   ASSERT(IsOriginalObject(object));
   word offset = 0;
-  if (target::CanLoadFromThread(object, &offset)) {
+  if (IsSameObject(compiler::NullObject(), object)) {
+    mov(dst, NULL_REG);
+  } else if (target::CanLoadFromThread(object, &offset)) {
     ldr(dst, Address(THR, offset));
   } else if (CanLoadFromObjectPool(object)) {
     const int32_t offset = target::ObjectPool::element_offset(
@@ -499,7 +501,9 @@ void Assembler::LoadUniqueObject(Register dst, const Object& object) {
 void Assembler::CompareObject(Register reg, const Object& object) {
   ASSERT(IsOriginalObject(object));
   word offset = 0;
-  if (target::CanLoadFromThread(object, &offset)) {
+  if (IsSameObject(compiler::NullObject(), object)) {
+    CompareRegisters(reg, NULL_REG);
+  } else if (target::CanLoadFromThread(object, &offset)) {
     ldr(TMP, Address(THR, offset));
     CompareRegisters(reg, TMP);
   } else if (CanLoadFromObjectPool(object)) {
@@ -1083,8 +1087,12 @@ void Assembler::StoreIntoObjectNoBarrier(Register object,
   ASSERT(IsOriginalObject(value));
   ASSERT(IsNotTemporaryScopedHandle(value));
   // No store buffer update.
-  LoadObject(TMP2, value);
-  str(TMP2, dest);
+  if (IsSameObject(compiler::NullObject(), value)) {
+    str(NULL_REG, dest);
+  } else {
+    LoadObject(TMP2, value);
+    str(TMP2, dest);
+  }
 }
 
 void Assembler::StoreIntoObjectOffsetNoBarrier(Register object,
@@ -1182,6 +1190,12 @@ void Assembler::RestoreCodePointer() {
   CheckCodePointer();
 }
 
+void Assembler::RestorePinnedRegisters() {
+  ldr(BARRIER_MASK,
+      compiler::Address(THR, target::Thread::write_barrier_mask_offset()));
+  ldr(NULL_REG, compiler::Address(THR, target::Thread::object_null_offset()));
+}
+
 void Assembler::CheckCodePointer() {
 #ifdef DEBUG
   if (!FLAG_check_code_pointer) {

runtime/vm/compiler/assembler/assembler_arm64.h

@@ -1515,8 +1515,12 @@ class Assembler : public AssemblerBase {
                                     uint32_t offset);
 
   void PushObject(const Object& object) {
-    LoadObject(TMP, object);
-    Push(TMP);
+    if (IsSameObject(compiler::NullObject(), object)) {
+      Push(NULL_REG);
+    } else {
+      LoadObject(TMP, object);
+      Push(TMP);
+    }
   }
   void PushImmediate(int64_t immediate) {
     LoadImmediate(TMP, immediate);
@@ -1556,6 +1560,10 @@ class Assembler : public AssemblerBase {
   void CheckCodePointer();
   void RestoreCodePointer();
 
+  // Restores the values of the registers that are blocked to cache some values
+  // e.g. BARRIER_MASK and NULL_REG.
+  void RestorePinnedRegisters();
+
   void EnterDartFrame(intptr_t frame_size, Register new_pp = kNoRegister);
   void EnterOsrFrame(intptr_t extra_size, Register new_pp = kNoRegister);
   void LeaveDartFrame(RestorePP restore_pp = kRestoreCallerPP);

runtime/vm/compiler/assembler/assembler_arm64_test.cc

@@ -2398,15 +2398,18 @@ static void EnterTestFrame(Assembler* assembler) {
   __ Push(CODE_REG);
   __ Push(THR);
   __ Push(BARRIER_MASK);
+  __ Push(NULL_REG);
   __ TagAndPushPP();
   __ ldr(CODE_REG, Address(R0, VMHandles::kOffsetOfRawPtrInHandle));
   __ mov(THR, R1);
   __ ldr(BARRIER_MASK, Address(THR, Thread::write_barrier_mask_offset()));
+  __ ldr(NULL_REG, Address(THR, Thread::object_null_offset()));
   __ LoadPoolPointer(PP);
 }
 
 static void LeaveTestFrame(Assembler* assembler) {
   __ PopAndUntagPP();
+  __ Pop(NULL_REG);
   __ Pop(BARRIER_MASK);
   __ Pop(THR);
   __ Pop(CODE_REG);
@@ -2481,6 +2484,39 @@ ASSEMBLER_TEST_RUN(LoadObjectNull, test) {
   EXPECT_EQ(Object::null(), test->InvokeWithCodeAndThread<RawObject*>());
 }
 
+// PushObject null.
+ASSEMBLER_TEST_GENERATE(PushObjectNull, assembler) {
+  __ SetupDartSP();
+  EnterTestFrame(assembler);
+  __ PushObject(Object::null_object());
+  __ Pop(R0);
+  LeaveTestFrame(assembler);
+  __ RestoreCSP();
+  __ ret();
+}
+
+ASSEMBLER_TEST_RUN(PushObjectNull, test) {
+  EXPECT_EQ(Object::null(), test->InvokeWithCodeAndThread<RawObject*>());
+}
+
+// CompareObject null.
+ASSEMBLER_TEST_GENERATE(CompareObjectNull, assembler) {
+  __ SetupDartSP();
+  EnterTestFrame(assembler);
+  __ LoadObject(R0, Object::bool_true());
+  __ LoadObject(R1, Object::bool_false());
+  __ ldr(R2, Address(THR, Thread::object_null_offset()));
+  __ CompareObject(R2, Object::null_object());
+  __ csel(R0, R0, R1, EQ);
+  LeaveTestFrame(assembler);
+  __ RestoreCSP();
+  __ ret();
+}
+
+ASSEMBLER_TEST_RUN(CompareObjectNull, test) {
+  EXPECT_EQ(Bool::True().raw(), test->InvokeWithCodeAndThread<RawObject*>());
+}
+
 ASSEMBLER_TEST_GENERATE(LoadObjectTrue, assembler) {
   __ SetupDartSP();
   EnterTestFrame(assembler);

runtime/vm/compiler/assembler/disassembler_arm64.cc

@@ -86,7 +86,7 @@ void ARM64Decoder::Print(const char* str) {
 static const char* reg_names[kNumberOfCpuRegisters] = {
     "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",  "r8", "r9",  "r10",
     "r11", "r12", "r13", "r14", "r15", "ip0", "ip1", "r18", "sp", "r20", "r21",
-    "r22", "r23", "r24", "r25", "thr", "pp",  "ctx", "fp",  "lr", "r31",
+    "nr",  "r23", "r24", "r25", "thr", "pp",  "ctx", "fp",  "lr", "r31",
 };
 
 // Print the register name according to the active name converter.

runtime/vm/compiler/backend/il_arm64.cc

@@ -940,10 +940,8 @@ void FfiCallInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
     __ blr(TMP);
   }
 
-  // Refresh write barrier mask.
-  __ ldr(BARRIER_MASK,
-         compiler::Address(
-             THR, compiler::target::Thread::write_barrier_mask_offset()));
+  // Refresh pinned registers values (inc. write barrier mask and null object).
+  __ RestorePinnedRegisters();
 
   // Although PP is a callee-saved register, it may have been moved by the GC.
   __ LeaveDartFrame(compiler::kRestoreCallerPP);
@@ -1102,10 +1100,8 @@ void NativeEntryInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   // Now that we have THR, we can set CSP.
   __ SetupCSPFromThread(THR);
 
-  // Refresh write barrier mask.
-  __ ldr(BARRIER_MASK,
-         compiler::Address(
-             THR, compiler::target::Thread::write_barrier_mask_offset()));
+  // Refresh pinned registers values (inc. write barrier mask and null object).
+  __ RestorePinnedRegisters();
 
   // Save the current VMTag on the stack.
   __ LoadFromOffset(R0, THR, compiler::target::Thread::vm_tag_offset());

runtime/vm/compiler/runtime_offsets_extracted.h

@@ -197,12 +197,12 @@ static constexpr dart::compiler::target::word
     Thread_array_write_barrier_entry_point_offset = 204;
 static constexpr dart::compiler::target::word Thread_async_stack_trace_offset =
     88;
-static constexpr dart::compiler::target::word
-    Thread_bootstrap_native_wrapper_entry_point_offset = 248;
 static constexpr dart::compiler::target::word
     Thread_auto_scope_native_wrapper_entry_point_offset = 256;
 static constexpr dart::compiler::target::word Thread_bool_false_offset = 112;
 static constexpr dart::compiler::target::word Thread_bool_true_offset = 108;
+static constexpr dart::compiler::target::word
+    Thread_bootstrap_native_wrapper_entry_point_offset = 248;
 static constexpr dart::compiler::target::word
     Thread_call_to_runtime_entry_point_offset = 208;
 static constexpr dart::compiler::target::word
@@ -553,12 +553,12 @@ static constexpr dart::compiler::target::word
     Thread_array_write_barrier_entry_point_offset = 392;
 static constexpr dart::compiler::target::word Thread_async_stack_trace_offset =
     176;
-static constexpr dart::compiler::target::word
-    Thread_bootstrap_native_wrapper_entry_point_offset = 480;
 static constexpr dart::compiler::target::word
     Thread_auto_scope_native_wrapper_entry_point_offset = 496;
 static constexpr dart::compiler::target::word Thread_bool_false_offset = 208;
 static constexpr dart::compiler::target::word Thread_bool_true_offset = 200;
+static constexpr dart::compiler::target::word
+    Thread_bootstrap_native_wrapper_entry_point_offset = 480;
 static constexpr dart::compiler::target::word
     Thread_call_to_runtime_entry_point_offset = 400;
 static constexpr dart::compiler::target::word
@@ -909,12 +909,12 @@ static constexpr dart::compiler::target::word
     Thread_array_write_barrier_entry_point_offset = 204;
 static constexpr dart::compiler::target::word Thread_async_stack_trace_offset =
     88;
-static constexpr dart::compiler::target::word
-    Thread_bootstrap_native_wrapper_entry_point_offset = 248;
 static constexpr dart::compiler::target::word
     Thread_auto_scope_native_wrapper_entry_point_offset = 256;
 static constexpr dart::compiler::target::word Thread_bool_false_offset = 112;
 static constexpr dart::compiler::target::word Thread_bool_true_offset = 108;
+static constexpr dart::compiler::target::word
+    Thread_bootstrap_native_wrapper_entry_point_offset = 248;
 static constexpr dart::compiler::target::word
     Thread_call_to_runtime_entry_point_offset = 208;
 static constexpr dart::compiler::target::word
@@ -1252,26 +1252,26 @@ static constexpr dart::compiler::target::word SubtypeTestCache_cache_offset = 8;
 static constexpr dart::compiler::target::word
     Thread_AllocateArray_entry_point_offset = 576;
 static constexpr dart::compiler::target::word Thread_active_exception_offset =
-    1352;
+    1344;
 static constexpr dart::compiler::target::word Thread_active_stacktrace_offset =
-    1360;
+    1352;
 static constexpr dart::compiler::target::word
     Thread_array_write_barrier_code_offset = 224;
 static constexpr dart::compiler::target::word
     Thread_array_write_barrier_entry_point_offset = 392;
 static constexpr dart::compiler::target::word Thread_async_stack_trace_offset =
     176;
-static constexpr dart::compiler::target::word
-    Thread_bootstrap_native_wrapper_entry_point_offset = 480;
 static constexpr dart::compiler::target::word
     Thread_auto_scope_native_wrapper_entry_point_offset = 496;
 static constexpr dart::compiler::target::word Thread_bool_false_offset = 208;
 static constexpr dart::compiler::target::word Thread_bool_true_offset = 200;
+static constexpr dart::compiler::target::word
+    Thread_bootstrap_native_wrapper_entry_point_offset = 480;
 static constexpr dart::compiler::target::word
     Thread_call_to_runtime_entry_point_offset = 400;
 static constexpr dart::compiler::target::word
     Thread_call_to_runtime_stub_offset = 264;
-static constexpr dart::compiler::target::word Thread_dart_stream_offset = 1424;
+static constexpr dart::compiler::target::word Thread_dart_stream_offset = 1416;
 static constexpr dart::compiler::target::word Thread_optimize_entry_offset =
     456;
 static constexpr dart::compiler::target::word Thread_optimize_stub_offset = 312;
@@ -1287,7 +1287,7 @@ static constexpr dart::compiler::target::word Thread_end_offset = 128;
 static constexpr dart::compiler::target::word
     Thread_enter_safepoint_stub_offset = 360;
 static constexpr dart::compiler::target::word Thread_execution_state_offset =
-    1392;
+    1384;
 static constexpr dart::compiler::target::word
     Thread_exit_safepoint_stub_offset = 368;
 static constexpr dart::compiler::target::word
@@ -1307,7 +1307,7 @@ static constexpr dart::compiler::target::word Thread_float_not_address_offset =
 static constexpr dart::compiler::target::word
     Thread_float_zerow_address_offset = 568;
 static constexpr dart::compiler::target::word Thread_global_object_pool_offset =
-    1368;
+    1360;
 static constexpr dart::compiler::target::word
     Thread_interpret_call_entry_point_offset = 504;
 static constexpr dart::compiler::target::word
@@ -1342,11 +1342,11 @@ static constexpr dart::compiler::target::word
 static constexpr dart::compiler::target::word Thread_object_null_offset = 192;
 static constexpr dart::compiler::target::word
     Thread_predefined_symbols_address_offset = 512;
-static constexpr dart::compiler::target::word Thread_resume_pc_offset = 1376;
+static constexpr dart::compiler::target::word Thread_resume_pc_offset = 1368;
 static constexpr dart::compiler::target::word
-    Thread_saved_shadow_call_stack_offset = 1384;
+    Thread_saved_shadow_call_stack_offset = 1376;
 static constexpr dart::compiler::target::word Thread_safepoint_state_offset =
-    1400;
+    1392;
 static constexpr dart::compiler::target::word
     Thread_slow_type_test_stub_offset = 344;
 static constexpr dart::compiler::target::word Thread_stack_limit_offset = 72;
@@ -1378,7 +1378,7 @@ static constexpr dart::compiler::target::word
 static constexpr dart::compiler::target::word Thread_write_barrier_mask_offset =
     96;
 static constexpr dart::compiler::target::word Thread_callback_code_offset =
-    1408;
+    1400;
 static constexpr dart::compiler::target::word TimelineStream_enabled_offset =
     16;
 static constexpr dart::compiler::target::word TwoByteString_data_offset = 16;
@@ -1414,7 +1414,7 @@ static constexpr dart::compiler::target::word
     Thread_write_barrier_wrappers_thread_offset[] = {
         1176, 1184, 1192, 1200, 1208, 1216, 1224, 1232, 1240, 1248, 1256,
         1264, 1272, 1280, 1288, -1,   -1,   -1,   -1,   1296, 1304, 1312,
-        1320, 1328, 1336, 1344, -1,   -1,   -1,   -1,   -1,   -1};
+        -1,   1320, 1328, 1336, -1,   -1,   -1,   -1,   -1,   -1};
 static constexpr dart::compiler::target::word Array_header_size = 24;
 static constexpr dart::compiler::target::word Context_header_size = 24;
 static constexpr dart::compiler::target::word Double_InstanceSize = 16;

runtime/vm/compiler/stub_code_compiler_arm64.cc

@@ -147,9 +147,8 @@ void StubCodeCompiler::GenerateCallToRuntimeStub(Assembler* assembler) {
   __ mov(SP, CSP);
   __ mov(CSP, R25);
 
-  // Refresh write barrier mask.
-  __ ldr(BARRIER_MASK,
-         Address(THR, target::Thread::write_barrier_mask_offset()));
+  // Refresh pinned registers values (inc. write barrier mask and null object).
+  __ RestorePinnedRegisters();
 
   // Retval is next to 1st argument.
   // Mark that the thread is executing Dart code.
@@ -624,9 +623,8 @@ static void GenerateCallNativeWithWrapperStub(Assembler* assembler,
   __ mov(SP, CSP);
   __ mov(CSP, R25);
 
-  // Refresh write barrier mask.
-  __ ldr(BARRIER_MASK,
-         Address(THR, target::Thread::write_barrier_mask_offset()));
+  // Refresh pinned registers values (inc. write barrier mask and null object).
+  __ RestorePinnedRegisters();
 
   // Mark that the thread is executing Dart code.
   __ LoadImmediate(R2, VMTag::kDartCompiledTagId);
@@ -1179,13 +1177,12 @@ void StubCodeCompiler::GenerateAllocateArrayStub(Assembler* assembler) {
   __ AddImmediate(R1, R0, target::Array::data_offset() - kHeapObjectTag);
   // R1: iterator which initially points to the start of the variable
   // data area to be initialized.
-  __ LoadObject(TMP, NullObject());
   Label loop, done;
   __ Bind(&loop);
   // TODO(cshapiro): StoreIntoObjectNoBarrier
   __ CompareRegisters(R1, R7);
   __ b(&done, CS);
-  __ str(TMP, Address(R1));  // Store if unsigned lower.
+  __ str(NULL_REG, Address(R1));  // Store if unsigned lower.
   __ AddImmediate(R1, target::kWordSize);
   __ b(&loop);  // Loop until R1 == R7.
   __ Bind(&done);
@@ -1257,9 +1254,8 @@ void StubCodeCompiler::GenerateInvokeDartCodeStub(Assembler* assembler) {
     __ mov(THR, R3);
   }
 
-  // Refresh write barrier mask.
-  __ ldr(BARRIER_MASK,
-         Address(THR, target::Thread::write_barrier_mask_offset()));
+  // Refresh pinned registers values (inc. write barrier mask and null object).
+  __ RestorePinnedRegisters();
 
   // Save the current VMTag on the stack.
   __ LoadFromOffset(R4, THR, target::Thread::vm_tag_offset());
@@ -1409,9 +1405,8 @@ void StubCodeCompiler::GenerateInvokeDartCodeFromBytecodeStub(
     __ mov(THR, R3);
   }
 
-  // Refresh write barrier mask.
-  __ ldr(BARRIER_MASK,
-         Address(THR, target::Thread::write_barrier_mask_offset()));
+  // Refresh pinned registers values (inc. write barrier mask and null object).
+  __ RestorePinnedRegisters();
 
   // Save the current VMTag on the stack.
   __ LoadFromOffset(R4, THR, target::Thread::vm_tag_offset());
@@ -2583,9 +2578,8 @@ void StubCodeCompiler::GenerateInterpretCallStub(Assembler* assembler) {
   __ mov(SP, CSP);
   __ mov(CSP, R25);
 
-  // Refresh write barrier mask.
-  __ ldr(BARRIER_MASK,
-         Address(THR, target::Thread::write_barrier_mask_offset()));
+  // Refresh pinned registers values (inc. write barrier mask and null object).
+  __ RestorePinnedRegisters();
 
   // Mark that the thread is executing Dart code.
   __ LoadImmediate(R2, VMTag::kDartCompiledTagId);
@@ -3072,8 +3066,8 @@ void StubCodeCompiler::GenerateJumpToFrameStub(Assembler* assembler) {
 #elif defined(USING_SHADOW_CALL_STACK)
 #error Unimplemented
 #endif
-  __ ldr(BARRIER_MASK,
-         Address(THR, target::Thread::write_barrier_mask_offset()));
+  // Refresh pinned registers values (inc. write barrier mask and null object).
+  __ RestorePinnedRegisters();
   // Set the tag.
   __ LoadImmediate(R2, VMTag::kDartCompiledTagId);
   __ StoreToOffset(R2, THR, target::Thread::vm_tag_offset());

runtime/vm/constants_arm64.cc

@@ -10,7 +10,7 @@ namespace arch_arm64 {
 const char* cpu_reg_names[kNumberOfCpuRegisters] = {
     "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",  "r8",  "r9",  "r10",
     "r11", "r12", "r13", "r14", "r15", "r16", "r17", "r18", "r19", "r20", "r21",
-    "r22", "r23", "r24", "ip0", "ip1", "pp",  "ctx", "fp",  "lr",  "r31",
+    "nr",  "r23", "r24", "ip0", "ip1", "pp",  "ctx", "fp",  "lr",  "r31",
 };
 
 const char* fpu_reg_names[kNumberOfFpuRegisters] = {

runtime/vm/constants_arm64.h

@@ -36,7 +36,7 @@ enum Register {
   R19 = 19,
   R20 = 20,
   R21 = 21,
-  R22 = 22,
+  R22 = 22,  // NULL_REG
   R23 = 23,
   R24 = 24,
   R25 = 25,
@@ -124,6 +124,7 @@ const Register THR = R26;           // Caches current thread in generated code.
 const Register CALLEE_SAVED_TEMP = R19;
 const Register CALLEE_SAVED_TEMP2 = R20;
 const Register BARRIER_MASK = R28;
+const Register NULL_REG = R22;  // Caches NullObject() value.
 
 // ABI for catch-clause entry point.
 const Register kExceptionObjectReg = R0;
@@ -173,9 +174,9 @@ const int kAbiPreservedFpuRegCount = 8;
 const intptr_t kReservedCpuRegisters =
     (1 << SPREG) |  // Dart SP
     (1 << FPREG) | (1 << TMP) | (1 << TMP2) | (1 << PP) | (1 << THR) |
-    (1 << LR) | (1 << BARRIER_MASK) | (1 << R31) |  // C++ SP
+    (1 << LR) | (1 << BARRIER_MASK) | (1 << NULL_REG) | (1 << R31) |  // C++ SP
     (1 << R18);
-constexpr intptr_t kNumberOfReservedCpuRegisters = 10;
+constexpr intptr_t kNumberOfReservedCpuRegisters = 11;
 // CPU registers available to Dart allocator.
 const RegList kDartAvailableCpuRegs =
     kAllCpuRegistersList & ~kReservedCpuRegisters;